gms | German Medical Science

The analysis of heart rate variability as non-invasive method is increasingly used in medicine. In the present paper we concentrate on the frequency-domain analysis of short-term heart rate variability measurements [Ref. 1].

The spectral density function is a commonly used tool whenanalyzing time series in the frequency domain. Using conventional nonparametric methods we obtain a direct spectral estimator by using an appropriate data taper (in order to reduce the bias of the periodogram estimator) and by then calculating the periodogram of the tapered data series. In order to get a smoother spectral estimator with better variance properties Welch [Ref. 2] suggested to split a time series into a number of overlapping blocks, compute a direct spectral estimate based on the data in each block and average the individual spectral estimates to form an overall one. This method is known as Welch's Overlapped Segment Averaging. Based on this approach Chave et al. [Ref. 3] proposed a method that is both, robust and data-adaptive.

We present this robust method applied to simulated and real data. As a special practical application we focus on actual heart rate variability measurements of diabetes patients.

The results of the proposed method are compared with those of classical non-robust methods and another robust one proposed by Martin and Thomson [Ref. 4]. The extension by Thomson's multitapering approach [Ref. 5] of the methodis also discussed.